CSP
CSP copied to clipboard
liuwei16
MR^2 can't achieve 4.62 on Caltech
I run this code in Caltech with paper's setting,but the best result tested by dbEval.m is 9.8. Anyone knows ?
Could you tell me what your threshold is？0.01 or other?My result of caltech is 30%.Thanks!
@cekcoco My paras is the same as the paper's. You can test the model that author published.
@cekcoco yes, other parameters is same as paper's,too. You can download directly.
which model produces the best result? 110 or 82? I download directly,but the result is wrong.Thanks.by the way,you result is 9.8,maybe the reason mentioned by the author is the version of opencv
@cekcoco 110. The version of opencv is same the author mentioned. Several days ago, I cried and laugh(哭笑不得). Now, I used pytorch code and change dataset.
@cekcoco 110. The version of opencv is same the author mentioned. Several days ago, I cried and laugh(哭笑不得). Now, I used pytorch code and change dataset.
Hi, how many images in your training set? The paper said use 40k, however, in the code the author is using 4k images. I also obtained the MR around 8-9%. Have you solved it?
@cekcoco 110. The version of opencv is same the author mentioned. Several days ago, I cried and laugh(哭笑不得). Now, I used pytorch code and change dataset.
Hi, how many images in your training set? The paper said use 40k, however, in the code the author is using 4k images. I also obtained the MR around 8-9%. Have you solved it?
No， I use code by pytorch and change the dataset..
hello, Is MR（mentioned in the paper ）the average of all the results or the best result? I'm confused. The result is roughly as shown below
59 | 5.208802 60 | 5.147443 61 | 5.000573 62 | 5.058751 63 | 5.197006 64 | 5.275083 65 | 5.172757 66 | 5.165324 67 | 5.098053 68 | 4.997828 69 | 4.778064 70 | 4.652336 71 | 4.799577
@cekcoco 110. The version of opencv is same the author mentioned. Several days ago, I cried and laugh(哭笑不得). Now, I used pytorch code and change dataset.
Hi, how many images in your training set? The paper said use 40k, however, in the code the author is using 4k images. I also obtained the MR around 8-9%. Have you solved it?
No， I use code by pytorch and change the dataset..
Hi, would you like to direct message me? My email is [email protected] and I am having the same issue with you. I also use the PyTorch code but unable to achieve a promising MR on the Caltech dataset. Maybe we can figure out together.
hello, Is MR（mentioned in the paper ）the average of all the results or the best result? I'm confused. The result is roughly as shown below
59 | 5.208802 60 | 5.147443 61 | 5.000573 62 | 5.058751 63 | 5.197006 64 | 5.275083 65 | 5.172757 66 | 5.165324 67 | 5.098053 68 | 4.997828 69 | 4.778064 70 | 4.652336 71 | 4.799577
the best
@lhs21 How did you get the result? Could you show me the details?
I just follow this code https://github.com/dominikandreas/CSP
@lhs21 How did you get the result? Could you show me the details?
I just follow this code https://github.com/dominikandreas/CSP
Thank you!
@lhs21 Can you give your test results file(in outputs/valresults/h/off/)?I want test it by Matlab. My qq:1136912015, Thank you!
@super-wcg Hello, have you solve this issue? I got the same miss rate (about 7.3%) when I just use the training model from the author, and the differential from the value is about 7.3-4.62=2.68%. Besides I use the CPU processing of the NMS, the other is the same as the original code.
When I train the Caltech code from https://github.com/dominikandreas/CSP, the records show the loss value is very high. I think that the value is the abnormal situation from cls_center, shown as below. But I have no idea to solve this issue. 0.146088 0.115322 0.009933 0.020833 0.098287 0.087741 0.002239 0.008307 0.071353 0.062760 0.001668 0.006925 0.081938 0.073539 0.001938 0.006461 0.073801 0.065529 0.001339 0.006932 0.072089 0.064259 0.001209 0.006621 0.067687 0.059619 0.001296 0.006772 0.064946 0.057368 0.001091 0.006488 0.061354 0.054006 0.001062 0.006286 0.061342 0.053803 0.001177 0.006362 0.061981 0.054710 0.001086 0.006185 0.058011 0.050864 0.001055 0.006091 0.055380 0.048392 0.001013 0.005975 0.059276 0.052155 0.001003 0.006118 0.051897 0.045055 0.000858 0.005984 0.052011 0.045238 0.000839 0.005935 0.055235 0.048431 0.000995 0.005809 0.053121 0.046175 0.000810 0.006136 0.054537 0.047586 0.001055 0.005896 0.050895 0.044234 0.000816 0.005845 0.051259 0.044498 0.000852 0.005909 0.050222 0.043484 0.000812 0.005926 0.049202 0.042674 0.000820 0.005707 0.047568 0.041040 0.000794 0.005733 0.045857 0.039506 0.000708 0.005643 0.049599 0.043301 0.000772 0.005527 0.046297 0.040108 0.000747 0.005442 0.045776 0.039386 0.000682 0.005708 0.047359 0.041113 0.000751 0.005495 0.044618 0.038325 0.000724 0.005569 0.044197 0.037804 0.000705 0.005688 0.043063 0.036738 0.000680 0.005645 0.044264 0.038139 0.000707 0.005418 0.041878 0.035751 0.000642 0.005485 0.044452 0.038305 0.000699 0.005448 0.042483 0.036077 0.000710 0.005695 0.044659 0.038537 0.000726 0.005396 0.041108 0.035030 0.000636 0.005442 0.040449 0.034342 0.000665 0.005442 0.040523 0.034651 0.000586 0.005286 0.040542 0.034645 0.000655 0.005242 0.039925 0.033994 0.000614 0.005317 0.040263 0.034208 0.000649 0.005406 0.039177 0.033443 0.000596 0.005138 0.037185 0.031399 0.000609 0.005177 0.041520 0.035570 0.000676 0.005274 0.039260 0.033446 0.000643 0.005171 0.038458 0.032674 0.000604 0.005180 0.038988 0.033193 0.000586 0.005209 0.038271 0.032380 0.000574 0.005317 0.037493 0.031569 0.000649 0.005276 0.038233 0.032616 0.000627 0.004990 0.037209 0.031312 0.000611 0.005286 0.038365 0.032619 0.000623 0.005123 0.036683 0.030863 0.000534 0.005286 0.036462 0.030927 0.000547 0.004988 0.035886 0.030214 0.000565 0.005108 0.037315 0.031443 0.000540 0.005332 0.035665 0.029956 0.000580 0.005128 0.036883 0.031298 0.000569 0.005016 0.036175 0.030355 0.000578 0.005241 0.034284 0.028385 0.000520 0.005379 0.034460 0.028736 0.000503 0.005221 0.034561 0.028825 0.000544 0.005192 0.035108 0.029526 0.000535 0.005047 0.033300 0.027814 0.000516 0.004970 0.035685 0.030010 0.000626 0.005049 0.034709 0.029089 0.000552 0.005067 0.032829 0.027182 0.000552 0.005095 0.034950 0.029222 0.000549 0.005179 0.035355 0.029635 0.000521 0.005199 0.033097 0.027575 0.000506 0.005016 0.034409 0.028776 0.000539 0.005094 0.034374 0.028723 0.000560 0.005091 0.033462 0.028019 0.000502 0.004941 0.031856 0.026466 0.000501 0.004889 0.034407 0.028888 0.000542 0.004977 0.033637 0.028005 0.000525 0.005106 0.032011 0.026455 0.000506 0.005049 0.032594 0.027016 0.000507 0.005072 0.031713 0.026079 0.000506 0.005128 0.032270 0.026873 0.000550 0.004847 0.033212 0.027798 0.000508 0.004907 0.033275 0.027806 0.000561 0.004908 0.031593 0.026026 0.000524 0.005043 0.033019 0.027241 0.000543 0.005235 0.031242 0.025833 0.000487 0.004922 0.033085 0.027534 0.000545 0.005007 0.032539 0.027065 0.000514 0.004960 0.031214 0.025757 0.000496 0.004960 0.031400 0.026045 0.000475 0.004880 0.030927 0.025401 0.000532 0.004995 0.031312 0.025998 0.000474 0.004839 0.031789 0.026305 0.000497 0.004988 0.031681 0.026299 0.000518 0.004863 0.029155 0.023938 0.000507 0.004711 0.028076 0.022962 0.000447 0.004667 0.030467 0.025188 0.000475 0.004803 0.031895 0.026434 0.000499 0.004961 0.030741 0.025282 0.000505 0.004954 0.030523 0.025147 0.000490 0.004887 0.031197 0.025962 0.000495 0.004740 0.028096 0.022867 0.000463 0.004767 0.029270 0.023728 0.000469 0.005072 0.030056 0.024727 0.000470 0.004859 0.029694 0.024355 0.000457 0.004883 0.029383 0.024017 0.000482 0.004884 0.029807 0.024387 0.000506 0.004914 0.030971 0.025645 0.000495 0.004831 0.031964 0.026566 0.000506 0.004892 0.030898 0.025541 0.000504 0.004853 0.029556 0.024205 0.000474 0.004878 0.028333 0.023096 0.000453 0.004784 0.028542 0.023254 0.000438 0.004850 0.028839 0.023586 0.000491 0.004762 0.029530 0.024289 0.000471 0.004770 0.030640 0.025408 0.000454 0.004779 0.030001 0.024737 0.000458 0.004806 0.029234 0.023925 0.000518 0.004791 0.028570 0.023454 0.000488 0.004627
Here is the version from the modules: Python:3.6 Keras:2.0.8 Tensorflow: 1.14.0 py-OpenCV: 3.4.2
If anyone knows what the reason is, please answer me, thanks!